Reducing the scheduling cost in event-driven simulation through component clustering

A method to reduce the scheduling overhead is proposed. Circuit elements that are likely to be activated simultaneously are grouped into clusters and are scheduled and evaluated simultaneously. This way, the number of entities involved in the scheduling, and therefore the scheduling overhead, is reduced. Different algorithms for partitioning the circuit into clusters are presented and the relationship between the cluster size and the simulation performance is studied. The algorithms presented were implemented for a switch-level simulator and tested for several large circuit descriptions. It is shown that with a carefully picked cluster size and partitioning algorithm, the scheduling overhead in the simulation can be significantly reduced and the simulation efficiency improved.<
>